Verba¶

What it is¶

Verba is an open-source Retrieval-Augmented Generation (RAG) application built on top of Weaviate. It provides a "Golden RAG" experience, focusing on simplicity and high-quality retrieval out of the box.

What problem it solves¶

It provides a user-friendly interface for building RAG applications, handling data ingestion, chunking, and querying with LLMs. It solves the complexity of setting up a complete RAG pipeline by providing a unified stack for experimentation and production use.

Where it fits in the stack¶

Category: Tool / Knowledge Management / RAG. It serves as the application layer on top of a vector database (Weaviate) to enable conversational search over private documents.

Typical use cases¶

Creating a personal knowledge base with AI search.
Question-answering over private document collections (PDF, Markdown, Text).
Testing different chunking and retrieval strategies.
Evaluating model performance (e.g., comparing Claude 4.8 vs GPT-5.5) on specific knowledge sets.

Strengths¶

Easy Setup: Reliable Docker-based deployment.
Multimodal Support: Built-in support for multiple data types (PDF, txt, etc.).
Native Weaviate Integration: Leverages Weaviate's advanced vector search, including hybrid search and reranking.
Model Flexibility: Supports latest frontier models like Llama 4 Maverick, Claude 4.8, and GPT-5.5.

Limitations¶

Ecosystem Lock-in: Closely tied to the Weaviate ecosystem.
Configuration Overhead: May require significant tuning for optimal performance with niche or extremely large datasets.
UI Constraints: The built-in frontend is optimized for specific RAG workflows and may not be easily customizable for all enterprise needs.

When to use it¶

When you want a production-ready RAG interface without building it from scratch.
For prototyping RAG workflows with Weaviate as the backend.
When you need a local-first RAG solution that can scale to cloud.

When not to use it¶

If you need a highly customized retrieval pipeline that departs significantly from Verba's modular architecture.
If you are already committed to a different vector database (e.g., Pinecone, Milvus) and do not wish to use Weaviate.

Getting started¶

Docker Deployment¶

The most reliable way to run Verba is via Docker Compose, which packages the frontend, backend, and Weaviate database.

git clone https://github.com/weaviate/Verba
cd Verba
# Set your API keys in the .env file
docker compose up -d

PIP Installation¶

pip install goldenverba
verba start

CLI examples¶

Verba provides a CLI for managing the application and data.

# Start the Verba server
verba start

# Import data into Verba from a specific path
verba import --path ./my_documents/

# Check the status of the Verba environment and connected components
verba status

API examples¶

Verba exposes a backend API that can be used to programmatically ingest data or query the RAG pipeline.

Query via Python¶

import requests

url = "http://localhost:8000/api/query"
payload = {
    "query": "How do I configure the OIDC middleware for Traefik?",
    "conversation_id": "optional-id",
    "model": "claude-4-8-opus-20260528"
}

response = requests.post(url, json=payload)
print(response.json()["answer"])

Weaviate — The vector database powering Verba.
Khoj — Alternative RAG assistant for personal notes and desktop search.
AnyType — Local-first P2P knowledge base.
RAG Pattern — Underlying architectural concept for retrieval augmentation.
Obsidian — Can be used as a primary data source for Verba.
LangChain — Framework often used to extend Verba's capabilities.
Ollama — Supported as a local inference backend for privacy-first RAG.
Model Context Protocol (MCP) — Standard protocol for connecting Verba to external tools.

Sources / references¶

Contribution Metadata¶

Last reviewed: 2026-06-28
Confidence: high